Parallel implementation of morphological operations on binary images using CUDA

Abstract

Morphology is a common technique used in image processing because it is a powerful tool with relatively low complexity. Albeit simple, morphological operations are typically time consuming due to the fact that the same operations are repeated on every pixel of an image. Since the processing of the pixels of an image is an embarrassingly-parallel process, the morphological operations can be carried out in parallel on Nvidia graphic cards using Compute Unified Device Architecture (CUDA). However, most of the existing CUDA work focuses on the morphological operations on grayscale images. For binary image, it can be represented in the form of a bitmap so that a 32-bit processor will be able to process 32 binary pixels concurrently. With the combination of the bitmap representation and van Herk/Gil-Werman (vHGW) algorithm, the performance of the proposed implementation in term of computation time improves significantly compared to the existing implementations.